Our goal is for AI systems to correctly identify and act according to their human user's objectives. Cooperative Inverse Reinforcement Learning (CIRL) formalizes this value alignment problem as a two-player game between a human and robot, in which only the human knows the parameters of the reward function: the robot needs to learn them as the interaction unfolds. Previous work showed that CIRL can be solved as a POMDP, but with an action space size exponential in the size of the reward parameter space. In this work, we exploit a specific property of CIRL---the human is a full information agent---to derive an optimality-preserving modification to the standard Bellman update; this reduces the complexity of the problem by an exponential factor and allows us to relax CIRL's assumption of human rationality. We apply this update to a variety of POMDP solvers and find that it enables us to scale CIRL to non-trivial problems, with larger reward parameter spaces, and larger action spaces for both robot and human. In solutions to these larger problems, the human exhibits pedagogic (teaching) behavior, while the robot interprets it as such and attains higher value for the human.

State-action value functions (i.e., Q-values) are ubiquitous in reinforcement learning (RL), giving rise to popular algorithms such as SARSA and Q-learning. We propose a new notion of action value defined by a Gaussian smoothed version of the expected Q-value. We show that such smoothed Q-values still satisfy a Bellman equation, making them learnable from experience sampled from an environment. Moreover, the gradients of expected reward with respect to the mean and covariance of a parameterized Gaussian policy can be recovered from the gradient and Hessian of the smoothed Q-value function. Based on these relationships, we develop new algorithms for training a Gaussian policy directly from a learned smoothed Q-value approximator. The approach is additionally amenable to proximal optimization by augmenting the objective with a penalty on KL-divergence from a previous policy. We find that the ability to learn both a mean and covariance during training leads to significantly improved results on standard continuous control benchmarks.

Recent work has shown local convergence of GAN training for absolutely continuous data and generator distributions. In this paper, we show that the requirement of absolute continuity is necessary: we describe a simple yet prototypical counterexample showing that in the more realistic case of distributions that are not absolutely continuous, unregularized GAN training is not always convergent. Furthermore, we discuss regularization strategies that were recently proposed to stabilize GAN training. Our analysis shows that GAN training with instance noise or zero-centered gradient penalties converges. On the other hand, we show that Wasserstein-GANs and WGAN-GP with a finite number of discriminator updates per generator update do not always converge to the equilibrium point. We discuss these results, leading us to a new explanation for the stability problems of GAN training. Based on our analysis, we extend our convergence results to more general GANs and prove local convergence for simplified gradient penalties even if the generator and data distribution lie on lower dimensional manifolds. We find these penalties to work well in practice and use them to learn high-resolution generative image models for a variety of datasets with little hyperparameter tuning.

The impact of continually evolving digital technologies and the proliferation of communications and content has now been widely acknowledged to be central to understanding our world. What is less acknowledged is that this is based on the successful arousing of curiosity both at the collective and individual levels. Advertisers, communication professionals and news editors are in constant competition to capture attention of the digital population perennially shifty and distracted. This paper, tries to understand how curiosity works in the digital world by attempting the first ever work done on quantifying human curiosity, basing itself on various theories drawn from humanities and social sciences. Curious communication pushes people to spot, read and click the message from their social feed or any other form of online presentation. Our approach focuses on measuring the strength of the stimulus to generate reader curiosity by using unsupervised and supervised machine learning algorithms, but is also informed by philosophical, psychological, neural and cognitive studies on this topic. Manually annotated news headlines - clickbaits - have been selected for the study, which are known to have drawn huge reader response. A binary classifier was developed based on human curiosity (unlike the work done so far using words and other linguistic features). Our classifier shows an accuracy of 97% . This work is part of the research in computational humanities on digital politics quantifying the emotions of curiosity and outrage on digital media.

Recent work in graph models has found that probabilistic hyperedge replacement grammars (HRGs) can be extracted from graphs and used to generate new random graphs with graph properties and substructures close to the original. In this paper, we show how to add latent variables to the model, trained using Expectation-Maximization, to generate still better graphs, that is, ones that generalize better to the test data. We evaluate the new method by separating training and test graphs, building the model on the former and measuring the likelihood of the latter, as a more stringent test of how well the model can generalize to new graphs. On this metric, we find that our latent-variable HRGs consistently outperform several existing graph models and provide interesting insights into the building blocks of real world networks.

This paper studies the potential of the return distribution for exploration in deterministic environments. We study network losses and propagation mechanisms for Gaussian, Categorical and Mixture of Gaussian distributions. Combined with exploration policies that leverage this return distribution, we solve, for example, a randomized Chain task of length 100, which has not been reported before when learning with neural networks.

Most Reading Comprehension methods limit themselves to queries which can be answered using a single sentence, paragraph, or document. Enabling models to combine disjoint pieces of textual evidence would extend the scope of machine comprehension methods, but currently there exist no resources to train and test this capability. We propose a novel task to encourage the development of models for text understanding across multiple documents and to investigate the limits of existing methods. In our task, a model learns to seek and combine evidence - effectively performing multi-hop (alias multi-step) inference. We devise a methodology to produce datasets for this task, given a collection of query-answer pairs and thematically linked documents. Two datasets from different domains are induced, and we identify potential pitfalls and devise circumvention strategies. We evaluate two previously proposed competitive models and find that one can integrate information across documents. However, both models struggle to select relevant information, as providing documents guaranteed to be relevant greatly improves their performance. While the models outperform several strong baselines, their best accuracy reaches 42.9% compared to human performance at 74.0% - leaving ample room for improvement.

Neural language models (NLMs) have recently gained a renewed interest by achieving state-of-the-art performance across many natural language processing (NLP) tasks. However, NLMs are very computationally demanding largely due to the computational cost of the softmax layer over a large vocabulary. We observe that, in decoding of many NLP tasks, only the probabilities of the top-K hypotheses need to be calculated preciously and K is often much smaller than the vocabulary size. This paper proposes a novel softmax layer approximation algorithm, called Fast Graph Decoder (FGD), which quickly identifies, for a given context, a set of K words that are most likely to occur according to a NLM. We demonstrate that FGD reduces the decoding time by an order of magnitude while attaining close to the full softmax baseline accuracy on neural machine translation and language modeling tasks. We also prove the theoretical guarantee on the softmax approximation quality.

Artificial neural networks tend to learn only what they need for a task. A manipulation of the training data can counter this phenomenon. In this paper, we study the effect of different alterations of the training data, which limit the amount and position of information that is available for the decision making. We analyze the accuracy and robustness against semantic and black box attacks on the networks that were trained on different training data modifications for the particular example of morphing attacks. A morphing attack is an attack on a biometric facial recognition system where the system is fooled to match two different individuals with the same synthetic face image. Such a synthetic image can be created by aligning and blending images of the two individuals that should be matched with this image.

In order for our computer systems to be more human-like, with a higher emotional quotient, they need to be able to process and understand intrinsic human language phenomena like humour. In this paper, we consider a subtype of humour - puns, which are a common type of wordplay-based jokes. In particular, we consider code-mixed puns which have become increasingly mainstream on social media, in informal conversations and advertisements and aim to build a system which can automatically identify the pun location and recover the target of such puns. We first study and classify code-mixed puns into two categories namely intra-sentential and intra-word, and then propose a four-step algorithm to recover the pun targets for puns belonging to the intra-sentential category. Our algorithm uses language models, and phonetic similarity-based features to get the desired results. We test our approach on a small set of code-mixed punning advertisements, and observe that our system is successfully able to recover the targets for 67% of the puns.